Another wild week, with Mistral AI coming out of nowhere to launch an open-source model that can be run on a laptop and, in early tests, seems to be close to GPT 3.5 levels.
Then, they raised $415M from incredible investors, including a16z, Lightspeed, and more.
I’m skeptical about whether they can catch up with OpenAI; however, it’s critical for there to be open-source alternatives. I believe there will be many opportunities for Mistral to outperform in specific tasks it’s trained on that are not as general as ChatGPT.
Rumors are spreading that GPT 4.5 is coming soon as a reaction to Mistral and Gemini. Exciting times ahead as competition heats up.
After initially postponing the launch of Gemini, a new LLM aiming to rival OpenAI's GPT-4, Google has launched a “lite” version called Gemini Pro.
Announced by CEO Sundar Pichai, Gemini will offer three versions: Gemini Nano for Android devices, Gemini Pro for AI services and powering Bard, and Gemini Ultra for data centers and enterprise applications.
Initial impressions of Gemini have been mixed, but it looks like the company has big plans in the pipeline. One of these is Project Ellmann: the concept of using LLMs like Gemini to ingest search results, spot patterns in a user’s photos, create a chatbot and “answer previously impossible questions” about a person’s life.
Elon Musk's Grok caused waves on social media after citing OpenAI's usage policy in its responses.
This raised questions about Grok's training data, as xAI's Igor Babuschkin admitted Grok inadvertently picked up ChatGPT outputs from the web during training.
However, experts like AI researcher Simon Willison suspect that Grok was fine-tuned using data from OpenAI models, a common practice in AI development. This situation has fueled the rivalry between OpenAI and xAI, especially given Musk's past criticisms of OpenAI.
The UK's Competition and Markets Authority (CMA) is reviewing the intensified relationship between Microsoft and OpenAI, following Microsoft's major investment in the AI firm.
This review aims to determine if their partnership resembles a merger, potentially raising competition concerns. The initial phase involves gathering comments, possibly leading to a formal investigation.
The focus is on whether the partnership leads to an acquisition of control or changes in control nature. This scrutiny follows recent governance changes at OpenAI, including CEO Sam Altman's brief role shift, amidst broader concerns about Big Tech's dominance in the AI market.
Google and DeepMind researchers, using the deep learning tool GNoME, have discovered 2.2 million new inorganic crystals, significantly expanding known stable structures.
Graph Networks for Materials Exploration (GNoME) identified 380,000 of these as highly stable, with 736 already experimentally realized. This breakthrough, equivalent to 800 years of material knowledge, has implications for superconductors and batteries.
Additionally, a second paper suggests using AI for an autonomous laboratory to accelerate new material synthesis, marking a major advancement in material science with potential medical and technological benefits.
AI made from living human brain cells performs speech recognition (Link)
Sports Illustrated boss Ross Levinsohn fired weeks after AI scandal (Link)
Meta’s AI for Ray-Ban smart glasses can identify objects and translate languages (Link)
EU agrees tough limits on police use of AI biometric surveillance (Link)
Runway introduces General World Models
Introducing General World Models.
We believe the next major advancement in AI will come from systems that understand the visual world and its dynamics, which is why we’re starting a new long-term research effort around general world models.